JP2007159131A - Admission control method and device for radio packet data services - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To accomplish a call admission control method and device extremely suitable for a radio system providing real-time services on a shared downlink. <P>SOLUTION: A call admission control technique takes into account both multiplexing and a multi-user diversity gain. The multi-user diversity gain is accurately determined by measuring resource allocation for each user so as to conveniently maximize not only a location dependent resource availability restriction but also a user accommodation capability under the quality of service (QoS). In a further embodiment, the call approval control technique is combined with delay-based scheduling effectively balancing system efficiency (channel usage rate) and a user expected value (e.g., QoS). A system realizing described call admission control and scheduling technique is capable of achieving conveniently efficient real-time services and keeps robustness for different loads varied by system dynamics and/or user mobility. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates generally to admission control technology, and more particularly to an admission control method and apparatus for a user of a wireless packet data system.

  Next generation wireless systems are expected to support on-demand applications such as real-time services. Real-time services on shared downlink channels need not only real-time schedulers but also efficient admission control, commonly called call admission control (CAC) (note that the term " Note that "call" is not limited to telephone calls or any particular media type).

  Admission control determines whether the system has resources to accommodate new users or calls such as streaming video or VolP requests without sacrificing the quality of service (QoS) of existing users. Judgment. The CAC function is particularly important for systems such as cellular radio systems where users are constantly entering and leaving the cell area while meeting the guaranteed service requirements for each user.

  Admission control is an essential part of the system and is one of the key elements that promote QoS in such a system. The admission control directly determines how efficient the system can be, i.e. how many real-time users the wireless system can support. For this reason, the approval control has a great influence on the satisfaction of the user and consequently the revenue of the system.

  There are many options for underlying packet level scheduling techniques, including Proportional Fairness (PF) method, Modified largest weighted delay first (MLWDF) method, and Exponential rule (Exp-Rule) algorithm.

  However, there are only a few real-time scheduling algorithms, in particular, real-time scheduling algorithms for shared channel cellular systems other than the MLWDF method and the Exp-Rule algorithm (see Non-Patent Documents 1 and 2).

  Both the MLWDF method and the Exp-Rule algorithm are weighted versions of the PF method where the load per user is equal to the head-of line (HOL) packet delay. Although the above methods show “throughput optimal”, they are not necessarily delay optimal.

  In wired network systems, earliest due date (EDD) and shortest time extinction (STE) algorithms have been shown to be optimal for minimizing average queuing delay and end-of-life violation packet loss, respectively. Has been.

On the other hand, as the simplest scheduling algorithm, FIFO queuing is known to minimize the maximum queuing delay. In the simple case where all incoming packets have the same end time D _s , the EDD algorithm and FIFO queuing results are equal. However, these techniques have low efficiency in cellular environments due to lack of channel recognition.

  For admission control, flow or user level approval in cellular downlink shared channel systems as opposed to many admission control technologies applicable to legacy circuit switched cellular systems where each user has a dedicated channel for output control. Few techniques to handle the control. At the flow level, only a few admission control techniques have been proposed for such systems (see Non-Patent Documents 3 and 4).

  However, the above approach struggles with inefficiencies due to ignoring multi-user diversity gain at the scheduler (or packet) level. In order to take advantage of multi-user diversity gain, it is necessary for the caller to choose the receiver with the best channel quality so as to maximize the spectral efficiency of the system.

  On the other hand, there are a number of channel-dependent opportunistic scheduling algorithms that use delay recognition or take advantage of such gains. However, these scheduling techniques assume a fixed number of users or a system with a static traffic load.

In particular, in cellular systems, in terms of accurately capturing the load of each user and the load of the entire system given the location dependence and user-specific channel quality, and the per-user QoS and opportunistic scheduler There are major challenges in characterizing the radio resources provided.
M. Andrews et al., “Providing Quality of Service over a Shared Wireless Link,” IEEE Commun. Mag., Pp. 15054 (Feb. 2001) S. Shakkottai and A. Stolyar, ”Scheduling Algorithms for Mixture of Real-time and NonReal-time Data in HDR,” in Proceedings 17th Int. TeletrafficCongress (ITC17) (Sept. 2001) T. Bonald and A. Proutiere, “Wireless Downlink Data Channels: User Performance and Cell Dimensioning,” Proceedings of ACM MOBICOM, pp.33952 (Sept. 2003) S. Das, H. Viswanathan, G. Rittenhouse, “Dynamic Load Balancing Through Coordinated Scheduling in Packet Data Systems,” IEEE Proceedings of INFOCOM, pp.78696 (April 2003)

  There are only a few real-time scheduling algorithms for shared channel cellular systems other than the MLWDF method and the Exp-Rule algorithm, and while the MLWDF method and Exp-Rule algorithm are “throughput optimal”, they are not necessarily delay optimal.

The simplest scheduling algorithm is known as FIFO queuing, which minimizes the maximum queuing delay, and in the simple case where all incoming packets have the same end time D _s , the EDD algorithm and FIFO queuing Although these results are equal, these approaches are less efficient in a cellular environment due to lack of channel awareness.

  For admission control, there are few techniques to handle flow or user level admission control in a cellular downlink shared channel system. At the flow level, only a few admission control techniques such as those disclosed in Non-Patent Documents 3 and 4 have been proposed, but these are efficient due to the neglect of multi-user diversity gain at the scheduler (or packet) level. I am struggling with evil. In order to take advantage of multi-user diversity gain, it is necessary for the caller to choose the receiver with the best channel quality so as to maximize the spectral efficiency of the system.

  Scheduling techniques in some channel-dependent opportunistic scheduling algorithms that take advantage of delay recognition or such gains are intended for systems with a fixed number of users or static traffic loads. When used in a cellular system, it accurately captures the load of each user and the load of the entire system given the location dependence and user-specific channel quality, and the QoS and optimism for each user There is a major problem in that the scheduler characterizes a given radio resource.

  The present invention has been made in view of the problems of the prior art as described above, and realizes a call admission control method and apparatus that are very suitable for a wireless system that provides a real-time service on a shared downlink. For the purpose.

The approval control method of the present invention includes:
Receiving scheduling decisions for communication resources;
Measuring an existing load of communication resources based on the scheduling decision;
Estimating the load of added new users,
Combining the measurement of the existing load and the estimation of the added load to determine a combined load;
Comparing the combined load with system capacity;
Approving the new user if the combined load does not exceed the system capacity;
including.

The admission control device of the present invention receives a scheduling decision for a communication resource, measures an existing load of the communication resource based on the scheduling decision, estimates a load of the added new user, A load estimator for determining a combined load by combining measurement and estimation of the added load;
The combined load is compared with a system capacity, and if the combined load does not exceed the system capacity, a determination unit that approves the new user;
including.

  In a first aspect, the present invention provides a call admission control technique for a wireless system having a shared downlink channel. The exemplary embodiment of the call admission control technique of the present invention considers both multiplexing and multi-user diversity gains that are accurately captured by online or real-time measurement of per-user resource allocation. The admission control technique of the present invention maximizes user capacity under constraints in QoS (eg, expected profile rate per user) and location-dependent resource availability constraints.

  As a further aspect of the present invention, the aforementioned call admission control is combined with delay-based scheduling, referred to herein as maximum cost deduction (MCD) scheduling. There are several exemplary embodiments of the scheduler that effectively balance between system efficiency (channel utilization) and user expectation (eg, QoS). A first scheduler, called real-time MCD (rt-MCD), minimizes the delay-derived cost function of raw packets to the smallest time scale (eg, slot level). A second scheduler implementation called non-real-time MCD (nrt-MCD) balances between real-time delay reduction and non-real-time (ie, a large temporal measure) per-user fairness.

  The call admission control and scheduling technique of the present invention effectively exploits multi-user diversity gain through accurate measurement of radio resources assigned to each user. They both combine high system efficiency and flow level QoS (eg, aggregate “goodput”, new incoming user blocking rate) and packet level QoS (eg, packet queuing delay or loss). Keep balance.

  A system embodying call admission control and scheduling according to the present invention is advantageous because it enables efficient real-time services and maintains robustness against different load scenarios that vary with system dynamics and / or user movement.

  The foregoing and other aspects and advantages of the present invention will become apparent to those of ordinary skill in the art by reference to the following detailed description and the accompanying drawings.

  Several performance metrics are cited here in describing the present invention. “Goodput for each user” means a transfer rate of a real-time packet in which packets are successfully distributed before the final deadline. Conversely, the loss rate of real-time packets means the proportion of packets dropped at the base station due to final expiration or delay violation. The average delay and jitter of successfully delivered real-time packets are also relevant metrics. The per user goodput and average delay and jitter metrics can also be measured for all users in route aggregation reflecting the overall system performance.

  Another metric is the call blocking rate for users that dynamically arrive at randomly distributed locations within the cell. From a system perspective, this metric also reflects the maximum number of users that can be approved under the packet level quality of service (QoS) constraints of the system. The performance fairness of complex real-time users, ie how the goodput of users approved by the system differs from their expected value of profile rate, indicates the robustness of the system. This metric is called a non-real time metric because it is not very delay sensitive.

  Each real-time user is presumed to specify strict QoS requirements at the packet level and nominal profile rates at the flow level. While strict QoS metrics such as packet delay and loss are monitored in each time slot, the profile rate matches the minimum or average streaming rate over a relatively long period of time. Because of volatile channel fading and intensive traffic / user arrival as well as high user mobility in the cellular environment, ensuring strict QoS is a difficult task if it can be done at all. . Therefore, practical systems attempt to maintain a balance between system efficiency, packet level QoS, and flow level performance.

  Another consideration is that for system robustness against abnormal conditions, such as whether a user's packet-level QoS approved during periods of stability or segmentation is still acceptable in the event of an abnormal condition, It is preferably robust under different load scenarios and drops in a predictable manner in the case of user mobility or system (load) dynamics.

An overview of the system used here and the QoS parameters are provided here.
Q _{k, s} (t) = {0,. . . , I,. . . , _{Nk, s} (t)}:
This is an unprocessed packet group in the FIFO queue of the cellular base station. Each packet is identified by p ⁱ _{k, s} (t), where, at time t, i is the packet index, s is the class ID, and k is the user ID of the intended recipient. Index i = 0 refers to the empty queue, i = 1 refers to the beginning of line (HOL) packet, and i = n _{k, s} (t) refers to the last packet in the FIFO queue.
Q _{k, s} (t):
A subset of Q _{k, s} (t) represents packets selected for transmission from the base station's FIFO queue at time t.
l ⁱ _{k, s} (t):
Length of packet p ⁱ _{k, s} (t) (eg bit length)
Δl ⁱ _{k, s} (t):
The length (for example, bit length) of the segment of the already transmitted packet p ⁱ _{k, s} (t).
m _{k, s} and m _k :
The average profile rate or minimum rate expectation (for example in kbps) of real-time flows for a single class s or all classes of user k, respectively. m _k is the sum of m _{k, s} in all active flows to user k.
D _s :
Delay amount of each packet from class s arriving at the cellular base station.
d ⁱ _{k, s} (t):
The queuing delay of packets p ⁱ _{k, s} (t) from that first arrival at the base station. This parameter includes the retransmission delay. Packet retransmission has a higher priority than the first transmission.
β _s :
This is the upper bound of packet loss (as part of a completely received packet) due to a deadline violation and is defined as:

Each real-time packet that violates the condition d ⁱ _{k, s} (t)? D _s is immediately removed from the buffer and is counted as a lost packet.
T _{k, s} (t) and T _k (t):
T _{k, s} (t) is an online measured average goodput in kbps for user k's class s.

Indicates the goodput for each user.
I _k (t):
A binary index that determines scheduling for user k at time t.
I _k (t) = 1 indicates that user k is scheduled at time t, and I _k (t) = 0 is the case where it is not scheduled.
B _k (t):
This parameter represents the average radio resource (eg bandwidth) allocated to user k at time t, where in an exemplary embodiment:

r _k (t) is the instantaneous channel speed of user k and t _l is the width of the smoothing window on a large time scale, eg t _l = 1,000 time slots.

  Using the above framework as background, several exemplary embodiments of the present invention will now be described in detail.

  FIG. 1 is a block diagram showing a configuration of an embodiment of a shared channel downlink radio communication system having a call admission control apparatus and a scheduler according to the present invention.

  The system shown in FIG. 1 schematically illustrates an exemplary shared channel downlink cellular system 100 having a call admission control (CAC) unit 110 and a scheduling unit 120.

  The CAC unit 110 and the scheduling unit 120 may be arranged, for example, in the base station 150 of the cellular system, but for example, together with other components of the mobile communication system including, in particular, a radio network controller (RNC). It can also be arranged. Further, the CAC unit 110 and the scheduling unit 120 need not be arranged together. For example, the CAC unit 110 may be arranged in the RNC and operate with a plurality of schedulers in a plurality of base stations.

  Exemplary operations of the CAC unit 110 and the scheduling unit 120 will be described in detail below.

  Whenever a new user 135 (shown as “new user i”) requests approval, the base station 150 provides services to the authorized users {1,..., K,. Will be. The CAC unit 110 determines whether to approve or reject each newly received user 135 in accordance with the present invention. In the context of wireless mobile communications, each “user” communicates with system 100 via a “mobile station” and, as such, means that the term is used interchangeably.

  The CAC block 110 includes a measurement-based load estimator 112 and a determination unit 114. The scheduling block 120 includes a packet classifier 122, a packet buffer 124, a maximum cost deduction (MCD) scheduler 126, and processing units 128 and 129.

In operation, incoming packets (ie, downstream packets p ⁱ _{k, s} (t) received by the system for transmission to authorized users 1... K) are received at base station 150 and passed to classifier 122. Provided. The classifier 122 determines the class s to which each packet belongs. Each user k can have 1 to S classes. Further, the classifier 122 maps the class identifier of the incoming packet to the class-specific parameters, that is, the profile rate m _s , the delay amount D _s , and the packet loss upper limit β _s . As will be described in more detail below, these class specific parameters are provided to the processor 128.

  Packets from each class s are queued in the packet buffer 124 for each user k. The classes are preferably sorted in order of increasing delay tolerance. In the exemplary embodiment, buffer 124 is a first-in first-out (FIFO) buffer with sufficient space to avoid frequent buffer overflows.

The buffer 124 determines queuing delays d ⁱ _{k, s} (t) from the buffered packets and the throughput history, and provides them to the processing unit 128. The processing unit 128 uses the information obtained from the classifier 122 and the buffer 124, as will be described in more detail below, depending on the scheduler implementation, the real-time delay-based weight function W _s (d ^l _{x, s} (t)), or non-real-time rate-based weight function

/ W _k (t)
Calculate

Each real-time flow is monitored at its network entrance by its profile rate m _{k, s} , and each incoming packet is, for example, S. Blake et al., “An Architecture for Differentiated Services,” Internet Engineering Task Force (IETF), Request. As described in for Comments 2475 (Dec. 1998), it is presumed that it is properly labeled with a Differentiated Services (DiflServ) code point. Real-time flows, such as video or audio streams, are usually carried by the IP / UDP / RTP protocol. For example, ITU G. The 729 coded voice over IP (VoIP) source produces 50 packets per second with a raw or compressed profile rate of approximately 24 kbps or 12 kbps.

On the other hand, ITU H. The H.263 encoded video source generates 25 frames per second of less time critical traffic at a profile rate of approximately 64 kbps. The end-to-end delay tolerance of each RT packet is in the range of about 150 milliseconds to 200 milliseconds. Finally, cellular access may be assigned a fixed delay amount (D _s ) of approximately 40 to 80 milliseconds, for example.

At each time slot t, the MCD scheduler 126, as will be described in more detail below, the QoS and queuing information of the outstanding packets, the profile rate {m _k } of all users and the instantaneous channel rate {r _k ( t)}, and “cost” per packet,

C ^l _{k, s} (t) / W _k (t)
Or based on C ^l _{k, s} (t), authorized users {1,. . . , K} group 125 selects the “best” user k ^* for packet transmission. As shown in FIG. 1, these parameters are provided from the processing unit 129 to the MCD scheduler 126. The shared channel downlink is dedicated to the best user k ^* (t) during timeslot t, i.e. it is that user whose waiting packet in buffer 124 is transmitted during that timeslot.

  By way of example only, time division multiplexed (TDM) channel access has been assumed, but the invention is not limited to any particular access scheme or specific system architecture.

As shown in FIG. 1, the load estimator 112 of the CAC block 110 receives the instantaneous scheduling decision group (I _k (t)) from the scheduler 126 and determines the instantaneous channel rate r _k (t) of each user. And determine the parameter B _k (t), the average radio resource allocation per user for each authorized user, according to equation (2) above. The load estimator 112 determines the group {B _k (t)}, the profile rate of authorized users (m _k ) and new users 135 (m _i ), the instantaneous channel rate {r _k (t)} of each user, And a new user's estimated instantaneous channel rate E [r _i (t)] to generate a normalized system load estimate. If the normalized load estimate is determined to be less than 1 by the determination unit 114, the new user is approved and a queue per user is formed for the new users in the buffer 124. Once approved, the new user's downstream packets are queued in the buffer 124 as described above. If the determination unit 114 determines that the normalized load estimate of the new user is 1 or more, the new user is denied approval. Some exemplary embodiments of the CAC block 110 and their operation are described in more detail below.

  The present invention is applicable to many systems, including third generation (3G) cellular systems, such as CDMA2000 High Data Rate (HDR) systems and WCDMA High Speed Data Packet Access (HSDPA) systems, It can be easily implemented without limitation.

 The present invention is described in, for example, MPEG4 or H.264 for mobile cellular users. While particularly advantageous for real-time services involving H.263 encoded video streams, the present invention is not limited to providing any particular service. Both systems employ a shared downlink channel to support multiple users for disparate expectations of real-time quality of service (QoS). The present invention advantageously provides for robust and efficient control of mobile users accessing such channels using disparate QoS expectations and recognition of multiple users' location-dependent channel conditions.

  An active call admission controller can operate with any scheduler that provides near instantaneous scheduling decisions. Several exemplary embodiments of the call admission controller and scheduler used in the above system will now be described in detail.

Channel-dependent, measurement-based CAC
In the system with K existing real-time users and newly arrived user i shown in FIG. 1, the following equations can be used to derive several call admission control algorithms.

In Equation (3), the numerical value “1” represents the normalized total channel capacity. G (K + 1) is a multi-user diversity gain that can be achieved with an opportunistic scheduler as described in more detail below. E [r _k (t)] and E [r _i (t)] are the average channel rates per user for existing user k and new user i, respectively. These parameters can be easily measured based on channel feedback.

and

Is the load per user applied to the shared channel by the existing user k and new user i, respectively. This metric for load takes into account both QoS expectations (eg profile rate) and user-specific channel quality (radio resource availability).

Various CAC algorithms can be derived from equation (3). For example, if the total system load is defined as the term L ₂ (t) and the system capacity is 1 in the above equation, the following CAC algorithm is derived, where user i is approved if the following equation is satisfied: .

  However, this algorithm may be prudent in some applications because the system load is determined as the “worst” case per user load.

Defining the total system load as term L ₁ (t) and system capacity as 1 in equation (3) gives a further CAC algorithm, represented here as CAC0, where the following equation is satisfied: User i is authorized.

  The CAC0 algorithm makes use of the multiplexing gain of a plurality of real-time users instead of the multiuser diversity gain.

  Another typical CAC algorithm that can be derived from Equation (3) is:

  This algorithm, shown here as CAC1, takes into account both multi-user diversity and multiplexing gain. However, the term G (K + 1) is actually difficult to handle analytically except in certain special cases. For example, when the Proportional Fairness (PF) scheduling algorithm is used and the normalized channel rate of all users

Are linear functions of SNR and are identical and independently distributed (IID: Identical and Independently Distributed)

It is shown that.

CAC1 algorithm can be actively used when set to.

  A more typical CAC algorithm can be expressed as: User i is approved when the following equation is satisfied:

  The algorithm represented by Equation (7) is denoted here as CAC2. As can be seen, CAC2 is the measured load per user,

All existing users {1,. . . , K} and the estimated load for new user i,

A pass / fail determination is made based on

As described above, B _k (t) represents the average radio resource (eg, bandwidth) allocated to user k at time t, where

t _l is the width of the smoothing window on a large time scale, eg t _l = 1,000 time slots.

By measuring B _k (t), the call admission controller can be executed according to the CAC2 algorithm, and the multi-user diversity gain inherent in the (wireless) bandwidth resource by the scheduler, if any, is accurate. Can get to. In particular, this measurement is independent of channel distribution and is not limited to any particular scheduling algorithm. In contrast, per-user goodput T _k (t) is constrained by limited traffic arrival in lightly loaded systems and thus does not accurately represent the bandwidth resource characteristics. Note that in the limited real-time flow of packet arrival, B _k (t)? T _k (t), where the equal sign applies only to users who have been left unprocessed for a long time.

  Furthermore, since CAC2 does not use G (K), which will be difficult to handle analytically, as explained, estimation becomes inaccurate.

  The test results described below show that CAC2 represents a good compromise between CAC0, which tends to be modest and CAC1, which tends to be aggressive.

Delay-Based Scheduling The system according to the present invention can be implemented with various schedulers, both real-time and non-real-time. Several schedulers that can be used in exemplary embodiments of the present invention are described in US patent application Ser. No. 11/276, filed Feb. 27, 2006, which is hereby incorporated by reference in its entirety. 381.

  In scheduling non-real time (NRT) data services, resource fairness and total system throughput are usually of primary concern. Existing NRT scheduling algorithms such as maxC / I and Proportional Fairness (PF) focus on channel state development under the assumption that the raw data is infinite. However, these algorithms cannot guarantee real-time packet delay or loss, resulting in poor goodput. This is due to the fact that the algorithm ignores dynamic traffic for raw data and that the algorithm does not pay enough attention to real-time packet delays in raw data. Resource fairness is a secondary problem for real-time services.

  Existing real-time schedulers for third generation (and later) cellular systems such as modified largest-weighted delay first (MLWDF) and Exponential Rule (Exp-Rule) schedulers provide real-time packet delay guarantees for multiuser diversity Effectively integrated with gain. Given the scheduler's ability to reduce packet loss, their goodput and robustness are limited by the scheduler's underlying mechanism for resource fairness. The scheduler described here provides real-time services that are comparable or better than before, and is robust in a wide range of system load scenarios.

  Real-time flow packet arrivals are non-regular, sudden and intensive. A good real-time scheduler should provide regular and timely service to unprocessed packets, because excessive delays in those packets cause deadline violations, resulting in packet loss and inadequate This results in raw data and low goodput.

  The downlink scheduler can achieve high system efficiency by effectively utilizing multi-user diversity gain. Unfortunately, for any user, the channel quality peak is almost inconsistent with the raw or queuing delay peak. Thus, the real-time scheduler preferably considers the balance between those users with good reception and users with time-critical flows. In each time slot, the scheduler should send the most expiring packets to the maximum channel capacity, which can generate a maximum cost deduction (MCD) from the system. In other words, the MCD scheduler pursues the following goals in each time slot:

Maximum {cost of delay derivation of dispatch packet} (8)
The system cost function can be defined as the total cost of all outstanding packets and can be expressed as:

Here, C ⁱ _{k, s} (t) represents the cost per packet of each queue packet p _{k, s} (t) at time t.

A successfully delivered real-time packet is a packet that has all its contents sent to the intended recipient of the packet before the end of the packet's final deadline. The cost of delaying larger or partially delivered packets is greater than the cost of delaying smaller or untransmitted packets. Thus, the per-packet cost function C ⁱ _{k, s} (t) is the total packet size l ⁱ _{k, s} (t), the arbitrary transmitted packet segment size Δl ⁱ _{k, s} (t) ), And a function of the queuing delay d ⁱ _{k, s} (t). The cost function C ⁱ _{k, s} (t) per packet should increase monotonically with l ⁱ _{k, s} (t) and Δl ⁱ _{k, s} (t). C ⁱ _{k, s} (t) is also a should increase d ⁱ _k, with monotonically _s (t), as d ⁱ _{k, s} (t) approaches the delay D _s, i.e., the packet is When dropped from the queue due to a delay violation, it approaches its maximum. Furthermore, C ⁱ _{k, s} (t) should differentiate packets of the same class s according to their delay, and should differentiate packets according to class.

  Taking the above features into account, an appropriate cost function per packet can be defined as:

Here, γ? 0 is a coefficient that weights the cost per packet of the length of the packet segment of the already transmitted segment Δl ⁱ _{k, s} (t), and W _s (d ⁱ _{k, s} (t) ) Is the unit cost per bit. The unit cost per bit, W _s (d ⁱ _{k, s} (t)), is a non-decreasing weighting function d ⁱ _{k, s} (t) of delay and is specific to each class s. This parameter is described in more detail below.

  As described, the present invention contemplates various schedulers. The first such scheduler is a real-time, maximum cost deduction (rt-MCD) scheduler with a delay-based weight function. The second scheduler is a non-real time MCD (nrt-MCD) scheduler with a rate-based weight function. Each of these will now be described in more detail.

  The rt-MCD scheduler first receives authorized users {1,. . . , K,. . . , K} group and {1,. . . , X,. . . Search for unprocessed real-time users, denoted as X}. If there are no unprocessed users, for example when the unprocessed data is empty or the system is lightly loaded, the scheduler looks as if all users have infinite unprocessed data and all packets have equal weight. Works as if it has.

However, if there are unprocessed users, the scheduler finds user x ^* (t) as follows due to their contribution to cost reduction.

The following conditions,

and

Where Q _{x, s} (t) is a packet selected for transmission from a queue of unprocessed packets (also called “(x, s) queue”) Q _{x, s} ( a subset of t).

A typical procedure for determining x ^* (t) according to equation (11) requires two optimization steps. For each user x, the scheduler first scans the unprocessed packet group Q _{x, s} (t) and the Q _{x, s} (t), (∀s) of each user x, within the user or between classes. Promotes cost deduction and determines the most “time critical” subset Q _{x, s} (t) that can be transmitted by the instantaneous channel rate r _x (t), ie, the scheduler packet from the mixed class as Select.

Packet segmentation is not possible, i.e., Δl ⁱ _{x, s} (t) = 0, C ⁱ _{x, s} (t) = W _s (d ⁱ _{x, s} (t)) l ⁱ _{x, s} (t) In this case, the packet selection problem of Equation (12) becomes a difficult NP-hard Knapsack problem. This problem can be solved with an approximation method, and the scheduler sorts packets from all classes / queues of user x according to the decreasing value of W _s (d ⁱ _{x, s} (t)) Select packets starting from the top of the list. The selection continues until the list is empty or the capacity is filled with the selected packet. The complexity of the approximation method is O (NlogN), where N is the total number of waiting packets for user x.

  However, if packet segmentation is possible, the scheduler will first decrease

Classify packets from all real-time classes in a single list. Starting from the top of the list, the scheduler selects packets or segments until the queue is empty or the channel capacity is full (with r _x (t) Δt bits). Note that the last selected “packet” may simply be a segment, ie, the packet is to be partially transmitted.

Given the selected packet { Q _{x, s} (t), (∀s)} for each user x, the scheduler proceeds and provides the maximum cost deduction based on the previous intra-user packet selection. Maximize the inter-user cost deduction by finding the best user x ^* (t) that may

The scheduling decision is I _x * (t) = 1, I _x (t) = 0 for all other _x .

With the rt-MCD scheduler as described, excellent QoS performance can be achieved using a linear or exponential delay-based weighting function W _s (d ⁱ _{x, s} (t)). Linear weight function, ie

Produces a linear per-packet cost function. Note that the delay is normalized with a class-specific delay amount D _s to obtain a class-independent weight for comparison purposes.

  An rt-MCD scheduler with a linear delay based weight function is shown here as an rt-MCD linear scheduler.

The exponential weight function, ie W _s (d) = ae ^{bd / Ds} , where a and b may be constant or change over time, but the packets are increasingly “time critical”, That is, in the case of d → D _s , it reflects a slight increase in unit delay cost that grows forever. In an exemplary embodiment, a = b = 1.

  The rt-MCD scheduler with an exponential delay-based weight function is denoted here as the rt-MCD-exp scheduler.

  As a special case,

,and

In this case, the rt-MCD-exp scheduler becomes the Exp-Rule scheduler. See S. Shakkottai and A. Stolyar, “Scheduling Algorithms for Mixture of Real-time and NonReal-time Data in HDR”, in Proceedings 17th Int. Teletraffic Congress (ITC 17) (Sept. 2001). Note that in EXP-Rule, d refers to the beginning of line (HOL) packet delay, that is, all packets in one queue are assumed to have the same delay, and each user k can accurately transmit one class of traffic. have.

  As explained, the second type of scheduler contemplated by the present invention is a non-real-time MCD (nrt-MCD) scheduler with a rate-based weight function. Real-time users

May also be interested in their long-term resource allocation compared to their profile rate expectation represented by where B _k (t) is determined according to equation (2) as described above. . From a system perspective, this metric reflects performance fairness, a non-real-time metric. Real-time schedulers can provide insufficient fairness over longer periods at the expense of long-term system efficiency to ensure fine-grained, short delays. To protect non-time critical real-time users (ie, their raw packets and current delays are far from the delay limit) from excessive raw packets and loss of efficiency, a rate-based weight function / W _k An MCD scheduler with (t) can be used in the system of the present invention. Such an nrt-MCD scheduler can be implemented as follows.

,and

Is assumed.

The nrt-MCD scheduler can be executed in the same procedure as in the case of the rt-MCD described above. While the intra-user or inter-class cost deduction for each user x finds the most “time critical” packet subset Q _{x, s} (t) as before, the inter-user cost deduction is optimal as Search for user k ^* (t).

Note that in this nrt-MCD scheduler, a simple linear function for W _s (d) was assumed in the cost function C ⁱ _{k, s} (t) per packet.

  Excellent QoS performance is a linear function

Can be realized with several different weighting functions. An nrt-MCD scheduler with such a weight function is denoted here as an nrt-MCD linear scheduler. Given infinite raw data, such a scheduler is equivalent to a weighted version of the PF algorithm where the weight is a user specific profile rate multiplied by the real-time packet delay.

Performance Evaluation To fully understand the improvements provided by the present invention over conventional methods, the combined performance of the CAC and scheduling algorithms of the present invention at both the flow and packet levels, such as user capacity and downlink throughput It is useful to evaluate the system capacity achieved for and the packet delay / loss metric for authorized users.

As an illustrative evaluation of typical implementation performance of the present invention, a CDMA / HDR downlink channel structure with a slot size of Δt = 1.667 milliseconds, ie, a scheduling frequency of 600 Hz, and a channel bandwidth of 1.25 MHz. Modeled. Average channel distribution follows a CDMA / HDR CDF function where each channel has a fast Rayleigh fading with 4 dB standard deviation and lognormal shadow fading, while the average supportable rate ranges from 153 kbps to 3.767 Mbps. Was assumed. In order to simplify the description without losing universality, the model was Assume only 263 real-time video streaming (S = 1) users and each user has only one flow. Each real-time flow was modeled as a Bernoulli process with a profile rate of m _k = 64 kbps. Given a packet length of 128 bytes in an HDR system, the average inter-packet arrival time is about 16 milliseconds.

Further, assuming a delay amount (D _s ) of 80 milliseconds, the HOL packet delay d ⁱ _{k, s} is used in Equation (10) which is a cost definition for all unprocessed packets of the same user k. This simplification actually reduces queue management costs without significantly different performance.

The model assumes that the CAC is performed on a single cell with a dynamic user population, users arriving and departing at a constant rate, and each user having a limited lifetime of 40 seconds. Yes. Furthermore, the average channel or bandwidth allocation is measured over a time window t _l of 1,000 time slots.

  For comparison, various system combinations of CAC0, CAC1, and CAC2 call admission controllers and Exp-Rule, rt-MCD, and nrt-MCD linear schedulers were modeled. It has been found that a system combination with an rt-MCD scheduler behaves similarly to a combination with an nrt-MCD linear scheduler.

  FIG. 2 shows the total packet loss rate for all combinations of CAC0, CAC1, or CAC2 call admission controller and Exp-Rule or nrt-MCD linear scheduler. The total packet loss rate reflects goodput. As shown in FIG. 2, a packet loss of 0 or close to 0 is obtained for the two combinations with CAC0 call admission. However, such a system has low goodput because the system accepts few users. Thus, the use of CAC0 provides full or near-perfect packet level QoS, while it causes a reduction in system efficiency or resource utilization. Furthermore, the close overlap of the two curves of CAC0 / Exp-Rule and CAC0 / nrt-MCD linear shown in FIG. 2 indicates that such performance is independent of the scheduler. Thereby, it can be confirmed that the expected value described above is the most conservative in CAC0 among the three CAC algorithms described above.

  FIG. 2 also confirms that CAC1 is the most aggressive of the three CAC algorithms, as shown by the high packet loss rate of the two curves CAC1 / Exp-Rule and CAC1 / nrt-MCD linear . Of the three CAC algorithms, CAC1 provides the highest total goodput (not shown), but at the expense of QoS as shown in FIG.

  CAC2 provides an excellent compromise between high goodput and low packet loss. Further, as shown in FIG. 2, the packet loss rate of a system implementing CAC2 does not vary significantly between the two implementations with the Exp-Rule scheduler and the nrt-MCD linear scheduler.

  FIG. 3 shows the average delay of successfully transmitted packets, each with an amount of delay of 80 milliseconds for various system configurations. Generally speaking, regardless of the scheduler, CAC0 and CAC1 respectively produce the lowest and highest average packet delays because they are too conservative and aggressive. As explained above, CAC2 operates at some point in between. FIG. 3 also shows that regardless of the CAC algorithm, a system with an nrt-MCD linear scheduler generally provides less delay than a system with an Exp-Rule scheduler. This is consistent with the expected value described above.

  FIG. 4 shows the average number of users authorized under different combinations of CAC and scheduling algorithms. The number of authorized users generally indicates the system capacity that is the inverse of the user blocking rate.

  As shown in FIG. 4, among the three CAC algorithms, CAC1 approves the highest number of users (but with higher packet delay and loss) and CAC0 approves the lowest number of users (and thus excellent QoS). But enjoy an inadequate goodput). In contrast, CAC2 recognizes a “correct” number of users that can be supported by the system in terms of both packet level QoS (eg, delay and loss) and system efficiency (eg, goodput).

  As shown in FIG. 4, the average number of authorized users for a system with CAC0 or CAC1 call authorization is practically independent of the scheduler. The CAC2 / Exp-Rule system can accommodate several more users than the CAC2 / nrt-MCD linear system.

  While representative drawings and specific embodiments of the present invention have been illustrated and illustrated, it is to be understood that the scope of the present invention is not limited to the specific embodiments described. Accordingly, the embodiments are to be regarded as illustrative rather than limiting, and those skilled in the art will recognize the scope of the invention as set forth in the claims below, and its structural and functional equivalents. It should be understood that variations of these embodiments can be made without departing from the invention.

1 schematically illustrates an exemplary embodiment of a shared channel downlink wireless communication system having a call admission controller and scheduler according to the present invention. 6 is a graph illustrating packet loss performance of various embodiments of a system according to the present invention. 6 is a graph illustrating packet delay performance of various embodiments of a system according to the present invention. 6 is a graph showing the performance of the number of authorized users of various embodiments of the system according to the invention.

Explanation of symbols

100 System 112 Load Estimator 114 Judgment Unit 122 Classifier 126 MCD Scheduler 125 User 128 Processing Unit 129 Processing Unit 135 New User 150 Base Station

Claims (19)

Receiving scheduling decisions for communication resources;
Measuring an existing load of communication resources based on the scheduling decision;
Estimating the load of added new users,
Combining the measurement of the existing load and the estimation of the added load to determine a combined load;
Comparing the combined load with system capacity;
Approving the new user if the combined load does not exceed the system capacity;
Including approval control method.   The method of claim 1, wherein the scheduling decision is substantially instantaneous.   The method of claim 1, wherein measuring the existing load comprises measuring radio resources allocated according to the scheduling decision.   The method of claim 1, wherein the communication resource is a shared downlink radio channel. The new user is approved if the following conditions are met:

Here, the new user K is the number of users k approved, m _k is profile factor of user k which has been approved, m _i is profile factor of new _{users, E [r i (t)} ] is the time t The estimated channel rate, B _k (t), is an average measurement of communication resources allocated to the authorized user k at time t.

Where r _k (t) is the channel rate of authorized user k at time t, and I _k (t) is the scheduling decision for the authorized user k at time t. The method of claim 5.   The method of claim 1, comprising rejecting the new user if the combined load exceeds system capacity.   The method of claim 1, wherein the method is performed at a cellular base station.   The method of claim 1, comprising generating the scheduling decision.   The method of claim 9, wherein the scheduling decision is generated according to a maximum cost deduction (MCD) algorithm.   The method of claim 10, wherein the MCD algorithm is a real-time MCD algorithm with a delay-based weight function.   The method of claim 11, wherein the delay-based weight function is a linear function.   The method of claim 11, wherein the delay-based weight function is an exponential function.   The method of claim 10, wherein the MCD algorithm is a non-real time MCD algorithm with a rate-based weight function.   The method of claim 14, wherein the rate-based weight function is a linear function.   The method of claim 9, wherein the scheduling decision is generated at a cellular base station.   The method of claim 1, wherein the method is performed at a radio network controller.   The method of claim 17, comprising generating the scheduling decision, wherein the scheduling decision is generated at a cellular base station. Receiving a scheduling decision of the communication resource, measuring an existing load of the communication resource based on the scheduling decision, estimating a load of the added new user, measuring the existing load and estimating the added load A load estimator that determines the combined load in combination;
The combined load is compared with a system capacity, and if the combined load does not exceed the system capacity, a determination unit that approves the new user;
Including approval controller.

Images